Groundwork

Groundwork

Requirements built on evidence, not opinions

Groundwork takes your team from customer insights to requirements, through hypotheses and experiments, so what you build rests on what customers showed you, not on gut feel.

Infusion prescribing · Intensive care

What clinicians said and did

Product analytics

Prescribing an infusion takes 15 minutes on average, three times longer than any other order.

Interview · ICU nurse

“I build the same noradrenaline infusion from scratch every shift.”

Interview · Ward physician

“Dilution, rate, volume… I type it all again for every patient.”

Draft hypothesis from 3 insights

When building is cheap, deciding what to build is the hard part

Coding agents are bringing the cost of building close to zero. Requirements are what's left. If they start from opinions, teams just ship the wrong thing faster. Groundwork makes sure every requirement starts from evidence.

Built on opinion

Built on evidence

The loudest voice in the room

What customers said and did

A hunch written as a ticket

A hypothesis tested before it ships

"Trust me, users want this"

A requirement with its evidence attached

From a customer's words to a shipped requirement.

Five connected steps take you from what customers said to requirements you can defend, with AI doing the heavy lifting at each one.

Customer insights

Keep what customers actually said

Capture interviews, tickets, surveys and analytics as small pieces of evidence. Paste raw notes and AI pulls out the quotes worth keeping.

  • Quotes kept in the customer's own words
  • Tagged by persona, source and product area
  • Linked to every decision they shape

Pulled from your interview notes

Product analyticsPrescribing

Infusion prescriptions take 15 minutes on average, three times longer than any other order.

InterviewICU nurse12 Nov

I build the same noradrenaline infusion from scratch every shift. Same dilution, same rate, every time.

DismissKeep quote
Support ticketCardiology ward

Can we save an infusion as a favourite for our unit?

Hypotheses

Turn a hunch into a belief you can test

Write who it's for, why it beats the alternatives and what success looks like. An AI coach challenges it, but never writes it for you.

  • Coached, never written for you
  • Success defined before anything is built
  • Riskiest assumption named up front

Coach has a question for you

DraftConfidence 40%

We believe discharge letters will go out on the day if medications carry over from the chart automatically.

TestingPrototype firstConfidence 45%

We believe infusion prescribing will drop below 5 minutes if a unit's most-used infusions are suggested, fully prefilled, on the prescribing screen.

DismissDesign experiment
ValidatedConfidence 82%

We believe clearer allergy alert wording will cut alert overrides.

Experiments

Decide what counts before you measure

Lock in success metrics and guardrails before an experiment starts, then record results as they come in, from a small bet to a full A/B test.

  • Targets locked before results arrive
  • Guardrails that must not get worse
  • Check-back dates for small bets

Results are in for 1 experiment

A/B test3 weeks

Medication list sorted by last given. Running on 4 wards, 860 prescriptions so far.

Prototype test10 cliniciansTarget < 5 min

Prefilled unit infusions: prescribing went from 15 minutes to 3. No wrong-dose errors.

DismissRecord verdict
Fake-door test1 week

Voice dictation for ward-round notes: 22% of physicians clicked.

Findings

Let the evidence deliver the verdict

Groundwork validates or invalidates the hypothesis and updates confidence. AI combines the hard numbers with the customer's voice. A no is a win too.

  • Validated or invalidated, never fudged
  • Numbers and quotes side by side
  • Confidence that moves with the evidence

New finding ready to review

InvalidatedScreen layout

Bigger dose buttons made no difference to prescribing time.

ValidatedConfidence 88%

Retyping the infusion, not the screen, was the slowdown. 15 minutes became 3, and clinicians said so too.

DismissDerive requirements
ValidatedAllergy alerts

Plain-language alert wording cut overrides by a third.

Requirements

Ship requirements that carry their receipts

Derive user stories and acceptance criteria from what you learned. Each one keeps its full trail back to the customer, all the way into your issue tracker.

  • Stories with acceptance criteria
  • Full evidence trail on every one
  • Send to Linear with the why attached

3 requirements derived from a finding

Should haveSolution

Clinicians can change any prefilled field before signing.

Must haveSolution5 links

As a prescribing clinician, I want my unit's most-used infusions suggested with everything prefilled, so that I can prescribe one in under 3 minutes.

DismissSend to Linear
Could haveStakeholder

Unit pharmacists can manage their unit's list of standard infusions.

Decide before you measure. Let the evidence decide.

Success criteria are set before an experiment starts. When results come in, Groundwork delivers the verdict and carries the learning forward, linked automatically.

  1. HypothesisPrefilled unit infusions cut prescribing time
  2. ExperimentPrototype test, 10 clinicians, target under 5 minutes
  3. Result15 minutes to 3 per prescription
  4. FindingRetyping the infusion, not the screen, was the slowdown
  5. RequirementSuggest the unit's most-used infusions, prefilled

Built for the whole product team

One shared source of truth, from the first interview to the last commit.

Prioritise with evidence, not opinions. Walk into roadmap reviews with the trail behind every bet.

Product managers

Keep research alive and connected to the decisions it shaped, instead of buried in a slide deck.

Designers

Requirements arrive with traceability built in, from customer quote to acceptance criteria.

Business analysts

Run a backlog stakeholders can trust, because every item shows why it's there.

Product owners

Understand the why behind every story without chasing anyone for context.

Engineers

See which insights changed what got built, and which are still waiting for a hypothesis.

Research leads

Fits the way you already work

Connect delivery, sign-in and your own AI key.

Linear

Send requirements with their evidence trail

Google sign-in

Your team is in with one click

Your own AI key

Free built-in AI, or your own key from Groq, OpenAI, Gemini, Anthropic or Ollama Cloud

Open source

Prefer your terminal? Groundwork is open source.

Your product evidence, in your repo, written by your own agent. Claude Code keeps it as linked Markdown files, with no account needed.

Ready to build what the evidence says?

Get started free

Frequently Asked Questions

What is Groundwork?

A product discovery tool that takes teams from customer insights to evidence-based requirements, through hypotheses, experiments and findings. Every requirement rests on evidence, not opinions, and keeps the trail that shows why.

Does the AI write my hypotheses for me?

No. The coach asks the hard questions (who is this for, what would prove you wrong, what's your riskiest assumption) but the belief stays in your words. AI does the tedious parts: pulling quotes out of notes, summarising findings and drafting requirements you then edit.

What happens to my data when I use AI?

Text is only sent to an AI vendor when someone in your workspace uses an AI feature: Groq by default, or the vendor your workspace chooses with its own key. Each workspace is visible to its members only. Please keep sensitive personal data, such as health records, out of a workspace.

Does it work with our issue tracker?

Requirements can be sent to Linear with their acceptance criteria and evidence trail attached, so engineers can see why a story exists.

Who is Groundwork for?

Product managers, designers, business analysts, product owners and the engineers who build what they decide.

How do I get started?

Sign in with Google or your email. You get a workspace straight away, and can invite your team when you're ready.